This repository contains the code and data necessary to reproduce the paper "Adversarial Classification: an Adversarial Risk Analysis approach". In particular it contains three different folders:
-
experimentSec3.4: This folder contains the code and data needed to reproduce the experiment on section 3.4 of the paper.
-
experimentsSec5: This folder contains the code and data needed to reproduce the first two experiments on section 5 of the paper.
-
2GWIExperiments: This folder contains the code and data needed to reproduce the last experiment on section 5 of the paper.
For most experiments, we used the Spambase Data Set from the UCI Machine learning repository, located in the data folder. This dataset consists of 4601 emails, out of which 1813 are spam (a 39.4% spam prevalence). For each email, the database contains information about 54 relevant words. We use the bag-of-words representation with binary features indicating the presence (1) or not (0) of such words, i.e. each email is assimilated with a vector of 0's and 1's of dimension 54.
For some experiments, we also used the Enron-Spam and Ling-Spam datasets. Documentation about these datasets available at https://www.cs.cmu.edu/~enron and http://csmining.org/index.php/ling-spam-datasets.html, respectively.